Lubuntu

irqbalance memory leak

16 7658
Hi, the problem shoud be fixt. You can get the code from here https://github.com/Irqbalance/irqbalance and compile it on your own. Ididnt try it jet.

tkaiser  
lordsandwurm replied at Mon May 4, 2015 16:22
Hi, the problem shoud be fixt. You can get the code from here https://github.com/Irqbalance/irqbalan ...

Ah, I thought the fix has been commited over a year ago: https://github.com/Irqbalance/irqbalance/commit/699deffdd881bf4759f3fcc4d6ee3e018e4c266e

But anyway: Why not just disabling irqbalanced? Isn't that the real problem: Pieces of software enabled by default and being not useful at all (on this specific platform. When I understood this and this correctly?)

vita  
I have a compiled latest version. There's no memory leak now. But as tkaiser said, irqbalance is not working with Allwinner SoC. All interrupt stay on CPU 0 even with latest version.

tkaiser  
Edited by tkaiser at Mon May 11, 2015 04:14
vita replied at Mon May 11, 2015 03:50
irqbalance is not working with Allwinner SoC. All interrupt stay on CPU 0 even with latest version.


So instead of simply disabling a piece of software that's not working on ARM devices lacking PCI/PCIe at all and not that good any more on modern SMP x86 systems, the 'solution' is to let the 'fixed' version just waste CPU cycles instead of eating up all the RAM too?

If you really care about performance related to interrupts simply stop irqbalance and maybe assign specific IRQs (USB, SATA or eth0 for example) to the second CPU core. The real solution would be to adopt the insight that irqbalance is useless on ARM and to remove it from the list of default packages on Banana distros (like the better ones already did).

vita  
Edited by vita at Mon May 11, 2015 14:16

I expressed it badly in my post . I wanted to say the same. Irqbalance is unnecessary. Personally, I also banned it.
So my post summary:
The irqbalance latest version has memoryleak solved, but does not work. So it is useless piece of software for bananapi

Even ARM says about IRQ balancing between Cortex core:
Load-balancing of interrupts across cores is not always the best solution
 Main considerations relate to cache migrations and contention, and their effect on performance  (note that the ARM MPCore™ architecture provides hardware optimizations and instructions to mitigate these)
 Example1: A device has multiple interrupts associated to it, and a unique control data structure
 Example2: Migrations overheads amplified by DFS
 Example3: It may be desirable to keep a CPU in idle state for longer so that the kernel will decide to put it in a deeper sleep state
 Example4: Interrupt balance detrimental for network latency

Source

tkaiser  
vita replied at Mon May 11, 2015 14:13
I expressed it badly in my post  . I wanted to say the same. Irqbalance is unnecessary. Personally ...

Thanks for the follow-up. So the solution for individual users of Lubuntu is
  1. sudo apt-get purge irqbalance
Copy the Code
and we've to hope that LeMaker doesn't include this package in their next OS releases. BTW: Even on the other side of the IT spectrum (not small ARM SBCs but big iron) today the key to performance is dedicated IRQ handling instead of random IRQ balancing (and then you further increase performance with eg. Receive Packet Steering and ensuring that only CPU cores that do not process ethN IRQs do the real packet handling)

Confirmed on my banana pro.   Out_of_memory killer starts killing processes after seven days or so, including irqbalanced.  I too, have a web cam connected along with a USB weather station.   I just purged irqbalanced sine most irqs were sent to cpu0 anyhow.

You have to log in before you can reply Login | Sign Up

Points Rules